1 · Pick a place
Click the map to move the pin. Dots are a sample of 2,500 block groups, coloured by recorded median value.
< $150k$150–300k$300–450k≥ $450k
2 · Describe the block group
3 · Estimate
–
At the dataset's $500,000 top-code: the true value could be higher.
5–95% range: –
Values are 1990 US dollars for a whole census block group, not one house.
About
- Model: scikit-learn gradient boosting (Huber loss, 400 trees, depth 5) on log value; 5% / 95% quantile models (150 trees, depth 4) give the range.
- Repeated 5-fold CV: see the results table, including a harsher spatial CV.
- Data: California Housing, R. K. Pace & R. Barry (1997), Sparse Spatial Autoregressions, Statistics & Probability Letters 33, 291–297; via StatLib and scikit-learn.
- Developed as part of my training with KF Property Network Pte Ltd (KFPN), a Knight Frank Singapore subsidiary, and submitted as my capstone for the NIELIT Ropar × IIT Ropar AI & ML programme.
Disclaimer: a personal educational project on public 1990 data. Not a valuation, not financial advice, and not endorsed by KFPN, Knight Frank or any other company.
Trilogy: I · Taiwan · II · Boston · III · California